Agents relying purely on imitation learning (IL) or reinforcement learning (RL) often struggle to avoid safety-critical errors during exploration. Existing RL approaches for environments such as chess require hundreds of thousands of episodes and substantial computational resources to converge. Unlike RL, IL enables efficient policy learning from expert demonstrations without requiring costly trial-and-error interaction, making it well-suited for complex domains like chess. In addition, IL can capture nuanced, human-aligned behaviors such as strategic patterns and stylistic preferences. However, IL-based agents may inherit biases from training data and lack safeguards against rare but critical mistakes, such as tactical blunders that can decisively impact game outcomes. In this work, we propose Oracle-Guided Soft Shielding (OGSS), a simple yet effective framework for safer move prediction, enabling safe exploration by learning a probabilistic safety model from oracle feedback in an imitation learning setting. In the domain of chess, we employ a two-model architecture: a move prediction model predicts strong moves from expert games, and a blunder model, trained on Stockfish evaluations, estimates tactical risk. At inference, the agent scores candidate moves using a utility function that combines move likelihood and blunder probability, enabling safe yet competitive game-play. When compared to other prominent methods such as action pruning, SafeDAgger, and uncertainty-based sampling, our results demonstrate that OGSS variants maintain a lower blunder rate even as the agent’s exploration ratio is increased by several folds, highlighting its ability to support broader exploration without compromising tactical soundness.

The potential applications for drones, especially those in manned areas or into non-segregated airspace, are currently not possible without the development and validation of certain key enabling technologies: "detect and avoid", "air traffic management" and "command and control (C2) link". SESAR JU identified that issue has a high impact on European innovation, which demands R&D investments and incentives for the convergence of shared technologies and markets. The COMP4DRONES project complements SESAR JU efforts with a particular focus on safe software and hardware drone architectures. COMP4DRONES will bear a holistically designed ecosystem ranging from application to electronic components. The ecosystem aims at supporting (1) efficient customization and incremental assurance of drone-embedded platforms, (2) safe autonomous decision making concerning individual or cooperative missions, (3) trustworthy drone-to-drone and drone-to-ground communications even in presence of malicious attackers and under the intrinsic platform constraints, and (4) agile and cost-effective design and assurance of drone modules and systems. Lead applications driving ecosystem development and benchmarking on the fields of transport, inspection, logistic, precision agriculture, parcel delivery, among others, will be produced.

International audience

Advanced Driver Assistance Systems (ADAS) and automated vehicle applications based on embedded sensors have become a reality today. As road vehicles increase its autonomy and the driver shares his role in the control loop, novel challenges on their dependability assessment arise. One key issue is that the notion of controllability becomes more complex when validating the robustness of the automated vehicle in the presence of faults. This paper presents a simulation-based fault injection approach aimed at finding acceptable controllability properties for the model-based design of control systems. We focus on determining the best fault models inserting exceptional conditions to accelerate the identification of specific areas for testing. In our work we performed fault injection method to find the most appropriate safety concepts, controllability properties and fault handling strategies at early design phases of lateral control functions based on the error in the Differential GPS signal.

The profile extension mechanism has permitted a rapid growth of the use of UML as a domain-specific modeling language. However, designing profiles typically falls into ad-hoc processes that often rely on domain-inappropriate primitives. One of the fundamental reasons is that profiles are specified on the same level of abstraction as the UML abstract syntax and consequently they narrow down the design space to an implementation level. In order to improvethis situation, some profile designers start from a "conceptual domain model" that states the domain ontology, and only then deal with finding out the profile extensions to support it. In spite of this, building truthfulness conceptual domain models and maintaining traceable mapping with the profile view is a bit of an art. In this paper, we propose to systematize the design of UML profiles built-upon conceptual domain models, by adopting a minimal setof framing rules. As these rules are defined on the basis of regularly occurring design patterns, domain models can be afterward checked for self-consistency and interactively transformed in stereotypes, tags and constraints